There is known an image generation device that maps an input image from a camera on a predetermined space model on a three-dimensional space, and generates a visual point conversion image, which is viewed from an arbitrary virtual visual point in the three-dimensional space while referring to the mapped space data (for example, refer to Japanese Patent Publication No. 3286306).
The image generation device disclosed in Patent Document 1 projects an image taken by a camera mounted on a vehicle onto a three-dimensional space model configured by a plurality of plane surfaces or curved surfaces that surround the vehicle. The image generation device generates a visual point conversion image using the image projected onto the space model, and displays the produced visual point conversion image to a driver. The visual point conversion image is an image of a combination of a road surface image, which virtually reflects a state of a road taken from directly above, and a horizontal image, which reflects a horizontal direction image. Thereby, the image generation device can relate, when the driver driving the vehicle looks at the visual point conversion image, an object in the visual point conversion image to an object actually existing outside the vehicle without giving an uncomfortable feeling.
The image generation device disclosed in Patent Document 1 generates a space model based on a feature point on a road surface that is image-taken by a camera, and corrects the space model in accordance with a movement of a vehicle (a change in a positional relationship between the vehicle and the feature point). A problem not considered is that, when generating or correcting a space model, an object in an overlapping area between image-taking ranges of two cameras disappears when projecting that object on a space model. Therefore, because the space model is not corrected in order to prevent such an image disappearance, a view point conversion image, which includes an object in an overlapping range, cannot be generated properly.